// Package laconic provides a lightweight research agent framework that uses

// a scratchpad-based architecture to keep prompts compact and efficient.

//

// Unlike traditional ReAct agents that append raw traces until context overflows,

// laconic maintains a compressed Scratchpad state and summarizes each iteration,

// making it practical on 4k/8k context models and low-cost LLM backends.

//

// # Architecture

//

// The agent operates in a Planner → Search → Synthesizer → Finalizer loop

// for the default scratchpad strategy:

//

// 1. The Planner examines the scratchpad and decides whether to answer or search.

// 2. If searching, the SearchProvider executes the query.

// 3. The Synthesizer compresses results into the scratchpad's knowledge state.

// 4. When ready (or max iterations reached), the Finalizer produces the answer.

//

// # Cost Tracking

//

// Every LLM call and search call can report a cost. LLMProvider.Generate returns

// an LLMResponse containing both text and cost. Search costs are configured via

// WithSearchCost. Agent.Answer returns a Result struct with the final answer and

// the total accumulated cost.

//

// # Basic Usage

//

// agent := laconic.New(

// laconic.WithPlannerModel(myLLM),

// laconic.WithSynthesizerModel(myLLM),

// laconic.WithSearchProvider(search.NewDuckDuckGo()),

// laconic.WithSearchCost(0.005),

// laconic.WithMaxIterations(5),

// )

//

// result, err := agent.Answer(ctx, "What is the capital of France?")

// fmt.Println(result.Answer)

// fmt.Printf("Cost: $%.4f\n", result.Cost)

//

// # Interfaces

//

// Implement LLMProvider to connect any language model:

//

// type LLMProvider interface {

// Generate(ctx context.Context, systemPrompt, userPrompt string) (LLMResponse, error)

// }

//

// LLMResponse carries the generated text and the cost of the call:

//

// type LLMResponse struct {

// Text string

// Cost float64

// }

//

// Implement SearchProvider to use any search backend:

//

// type SearchProvider interface {

// Search(ctx context.Context, query string) ([]SearchResult, error)

// }

//

// Strategy options include the default scratchpad loop and the graph-based

// "graph-reader" strategy.

// See the examples/basic directory for a complete example.

package laconic